Frequently Asked Questions
Common questions about how Maindex works, what things cost, and how your data is handled.
Why not use my LLM's built-in memory?
Built-in LLM memory is typically a flat list of text notes with no structure, no search beyond keyword matching, and no way to export or move your data to another service. If you switch providers, your memories stay behind. If your note list grows, discoverability drops because the LLM has no semantic search, no collections, no tags, and no way to surface the right context at the right time.
Maindex gives you a structured, searchable memory graph accessible through both an agent-facing MCP interface and a human-facing dashboard. Your data is portable across every platform that supports MCP, and you can build your own integrations with the REST API.
Why not use markdown or text files with my desktop app?
Local files work okay until the memory pool grows. Basic text search scales poorly; every query has to scan more content, and your agent burns context window and time on retrieval instead of doing real work. There is no semantic understanding, so finding related ideas across files means the agent is guessing at filenames and grepping through directories.
Maindex offers vectorized semantic search that stays fast regardless of how large your memory graph gets. Your data is stored centrally and accessible from any platform. Draft a spec on your phone, then pull those same memories into your IDE to execute the plan. No syncing folders, no copy-pasting between devices.
What is the Smart service?
Smart is the streamlined MCP interface with four tools: keep, recall, update, and forget. It is designed for agents and platforms that prefer a minimal, opinionated tool surface. Behind the scenes, Smart routing handles retrieval strategy, ambiguity resolution, and content shaping so the agent doesn't have to.
Both Smart and Expert are included with every Maindex plan at no additional cost, offering two paths to the same data.
What is the Expert service?
Expert is the full-control MCP interface with 14 tools: granular CRUD, filtered search, collection management, typed associations, bulk operations, and supersession chains. It gives agents direct access to every capability in the memory graph.
Both services are included with every Maindex plan at no additional cost, offering two helpful paths to the same data.
Should I use the Smart or Expert MCP?
Both are included, both read and write the same memory graph, and you can switch at any time or use both simultaneously. Use Smart when you want the agent to send high-level intents and let the pipeline decide. Use Expert when the agent needs direct, granular control over memory operations.
Some MCP platforms work best with a smaller tool surface, which is why Smart exists. There is no penalty for choosing one over the other, and memories created in one are immediately visible in the other.
Teams & Privacy
What are Teams?
Teams let several people share memory collections that all of their agents can read and write. You create a team, invite people by email, and give each of them a role. Everyone connects through the same MCP setup they already use. Team memory shows up alongside their personal memory, clearly labelled.
Can I join a team on any plan?
Yes. Every paid plan includes team access; you can join unlimited teams. Creating your own team requires Galaxy Brain. Everyone can pay for themselves, or one person can cover the whole team on one invoice. Same plan prices either way. Plan limits → · Teams guide →
Who can create a team?
Anyone on Galaxy Brain, who can create unlimited teams. Pending invites count toward the member limit. Creating a team →
Can my team's manager see my personal memories?
No. Never, under any circumstance. Personal memories are unreachable from team scope. A manager looking one up gets "not found," and we don't even confirm it exists. It's enforced by database row-level security, not by an application check or a setting someone could change. Managers see team activity and daily read/write totals on the team's collections, and that's it. The boundary, in detail →
How does team namespacing work?
Team and personal memory are separate scopes. A write lands in personal memory unless you name the team explicitly.
On write tools, pass team (for example team: "acme"). You can also use a prefixed collection slug like acme:runbooks when the collection name alone would be ambiguous.
IDs follow the same rule: acme:mem-4f is team memory; bare mem-4f is personal. Omit team on reads to search personal memory plus every team you belong to, each result labelled by scope. Using Teams with your agents →
How does automated review work?
Writes to team collections are screened on our servers with automated checks and agentic review before the team can see them. We look for prompt injection, secrets and API keys, injected instructions, and profanity. Most writes pass and publish immediately. Anything flagged is held for a manager to approve, reject, or redact, and the writing agent is told what happened on its next call. Screening cannot be turned off by an MCP client. What gets screened →
What do managers see in the audit log?
Every team action, in an append-only record: team creation and settings changes, invites, role changes, ownership transfer, freezes, collection changes, memories copied in or out, review verdicts, and shadow-copy actions. It's queryable by actor, action, and date, and covers team activity only. Nothing a member does in their personal account appears in it. Audit and analytics →
My company pays for my account: what happens if I leave?
Your personal graph stays attached to your email. Team contributions stay with the team. Resubscribe on your own plan and your memories are there. Your login and personal memories are never tied to your employer's billing group.
Can we pay for everyone on one invoice?
Yes. One person creates a billing group and covers everyone on the team. Seats can be added and removed as the team changes. Plan prices are the same as retail; Group Billing is about coordination, not discounts.
What happens to team memories if I leave, or if the team is deleted?
Memories you wrote to a team belong to the team, so they stay when you leave. Your access is revoked; the content remains. If a team is deleted, the owner starts a deferred deletion with a grace period during which it can be cancelled. Managers can export any team collection to JSON at any time before that. Your personal memories are never affected by anything that happens to a team. Leaving and deleting →
Usage & Data
What are synapses?
A synapse is a single unit of memory usage. Every time an agent or API call reads from or writes to your memory graph, the processing is measured in synapses. Your plan includes a generous monthly synapse quota that resets each billing period. You can see your current usage and reset date on the Usage & Billing tab in your dashboard.
What is a collection?
A collection is a named group of related memories. Use them to organize your knowledge graph by project, domain, or any structure that makes sense to you. A memory can belong to more than one collection, so you can slice your graph in multiple ways without duplicating anything. Agents can filter recalls to a specific collection so they only see what is relevant.
What are locked collections?
A locked collection is protected by a passphrase. Memories inside a locked collection are hidden from all sessions and API calls until you explicitly unlock them. This lets you compartmentalize information (work vs personal, private notes, client data) in the same memory graph without exposing it to every agent session. You can selectively unlock individual collections when you need them.
The locked collection management page uses an additional authentication layer that prevents agents from attempting to bypass the lock and access protected data without your authorization.
Can I export my data?
Yes. Your dashboard has a Data Portability section under Settings with two export options:
- Full Export (JSON): memories, tags, collections, links, revision history, and conversation references. Everything.
- Memories JSON: just your memories in a clean JSON format.
You can also import a Full Export JSON back into a new account. Your data is yours on any active account with dashboard access.
How does data export work?
You own your thoughts and we honor that with full portability. Export everything (memories, links, collections, tags, revision history) from your dashboard while your account is active. Import it into a new account or keep a local backup. No walled garden, no gilded cage.
Do you train models on my data?
No. Maindex does not train models on your data. We never use your memories, metadata, tags, links, or any part of your knowledge graph for model training, analytics, advertising, or any purpose other than serving your data back to you.
Is my data private?
Every Maindex account is a fully isolated tenant. Your data is never shared, mixed, or accessible by other users. All data is encrypted at rest with AES-256 and in transit with TLS 1.2+. Locked collections add an additional passphrase-based access control layer on top of that.
Authentication is OAuth-first via WorkOS, so Maindex stores no passwords. API keys with per-key revocation are available for headless environments. Step-up re-authentication is required for destructive operations like account deletion or data export.